CN111803957A

CN111803957A - Player prediction method and device for online game, computer equipment and medium

Info

Publication number: CN111803957A
Application number: CN202010694180.3A
Authority: CN
Inventors: 瞿曼湖; 赵世玮; 吴润泽; 张怡婷; 陶建容; 沈旭东; 范长杰; 胡志鹏
Original assignee: Netease Hangzhou Network Co Ltd
Current assignee: Netease Hangzhou Network Co Ltd
Priority date: 2020-07-17
Filing date: 2020-07-17
Publication date: 2020-10-23
Anticipated expiration: 2040-07-17
Also published as: CN111803957B

Abstract

The application provides a player prediction method, a device, a computer device and a medium of a network game, wherein the method comprises the following steps: acquiring first general attribute data in a target game and general game consumption data of a target player; and inputting the first general attribute data and the general game consumption data into a trained prediction model to obtain the churn probability and the payment probability of the target player in the target game. According to the embodiment of the application, the player prediction of different online games is realized through the two general data, namely the first general attribute data and the general game consumption data. The first general attribute data and the general game consumption data are input into a trained prediction model at the same time, so that the loss probability and the payment probability of the target player can be predicted by simultaneous calculation, the loss probability and the payment probability of the target player can be predicted at the same time, and the prediction efficiency of the target player is improved.

Description

Player prediction method and device for online game, computer equipment and medium

Technical Field

The present application relates to the field of game prediction, and in particular, to a player prediction method, apparatus, computer device and medium for a network game.

Background

In the current society, the continuous change of science and technology can be suitable for the rapid development of network games in intelligent terminal equipment, the network games enter the lives of people, players can play and communicate with partners of every party in the network games, the distance between people is shortened, and a more convenient friend making mode is provided for the players.

In terms of game developers of network games, how to retain players of network games is a problem that needs to be considered by the game developers. Therefore, the game developer can predict the player according to the data after the network game is run, and reduce the loss of the player according to the prediction result.

Disclosure of Invention

The applicant finds that, in the long-term development process, the scheme in the related art has at least the following disadvantages: on the one hand, the player churn probability or payout probability can only be predicted singularly; on the one hand, the expansibility is poor, and different games cannot be well adapted. In view of the above, an object of the present application is to provide a player prediction method, device, computer device and medium for online games, which are used to solve the problems in the prior art that the churn probability and the payment probability cannot be predicted for one game at the same time, and that different games cannot be well adapted.

In a first aspect, an embodiment of the present application provides a player prediction method for a network game, including:

acquiring first general attribute data of a target player in a target game and general game consumption data of the target player;

and inputting the first general attribute data and the general game consumption data into a trained prediction model to obtain the churn probability and the payment probability of the target player in the target game.

Optionally, the first general attribute data includes any one or more of the following data:

occupation, gender, player level, VIP level, online duration, character creation time, character server, character experience, daily charge amount, accumulated charge amount, daily consumption amount, accumulated consumption amount, and game character ID number of the target player.

Optionally, the general game consumption data includes any one or more of the following data:

the method comprises the following steps of purchasing prop name data of the props, purchasing money data of the props, purchasing time data of the props and purchasing time interval data of the props.

Optionally, inputting the first general attribute data and the general game consumption data into a trained prediction model to obtain the churn probability and the payout probability of the target player in the target game, including:

inputting the occupation, the gender, the online time and the game role ID number in the first general attribute data into a first characteristic evaluation model to obtain a first evaluation result related to the first general attribute data;

inputting prop name data for purchasing props, sum data for purchasing props, time data for purchasing props and time interval data for purchasing props in the general game consumption data into different evaluation models respectively to obtain evaluation results corresponding to each data;

and calculating the churn probability and the payment probability of the target player in the target game according to the evaluation results corresponding to the prop name data of the purchased props, the sum data of the purchased props, the time data of the purchased props and the time interval data of the purchased props and the first evaluation result corresponding to the first general attribute data.

and inputting the first general attribute data, the second general attribute data of the game role and the general game consumption data into a trained prediction model to obtain the churn probability and the payment probability of the target player in the target game.

Optionally, the second general attribute data includes any one or more of the following data:

name data of a game character, equipment data of a game character, and skill data of a game character.

Optionally, the prediction model is obtained by training through the following steps:

acquiring a training sample set; the sample set comprises at least one training sample, and each training sample comprises first training data and second training data; the first training data comprises first generic attribute data of the player and generic game consumption data of the player; the second training data includes attrition labels and bet labels for the player;

and aiming at each training sample, simultaneously inputting the first training data and the second training data in the training sample to a prediction model to be trained, and training the prediction model to be trained.

Optionally, for each training sample, the first training data and the second training data in the training sample are simultaneously input to the prediction model to be trained, and the training of the prediction model to be trained includes:

and for each training sample, according to a comparison result between the calculation result of the first training data and the second training data, adjusting the weight value of each feature in the prediction model to be trained so as to train the prediction model to be trained until the comparison result between the calculation result of the positive sample and the negative sample reaches preset precision.

Optionally, the method further includes:

and determining to push a message about the target game to the target player according to the attrition probability and the payment probability.

Optionally, the message about the target game includes any one or more of the following messages:

a coupon strategy regarding the target game and an improvement strategy regarding the target game.

Optionally, the offer policy includes any one or more of the following policies:

providing lottery activity, providing game privileges, upscaling property discounts.

Optionally, the improvement policy includes any one or more of the following policies:

optimizing game pages, increasing game level and improving game fluency speed.

Optionally, determining to push a message about the target game to the target player according to the churn probability and the payment probability includes:

if the churn probability of the target player is smaller than a first churn probability preset value and the payment probability is smaller than a first payment probability preset value, determining to push a preferential strategy to the target player;

if the churn probability of the target player is larger than a first churn probability preset value and smaller than a second churn probability preset value, and the payment probability is smaller than a first payment probability preset value and smaller than a second payment probability preset value, determining that two preferential strategies are simultaneously pushed to the target player;

and if the churn probability of the target player is greater than a second churn probability preset value and the payment probability is greater than a second payment probability preset value, determining to simultaneously push three preferential strategies to the target player.

and if the churn probability of the target player is greater than a third churn probability preset value and the payment probability is greater than a third payment probability preset value, determining to push a discount strategy about the target game and an improvement strategy about the target game to the target player.

Optionally, the obtaining first general attribute data of a target player in a target game and general game consumption data of the target player includes:

acquiring log data corresponding to the target player from a database of the target game;

and screening out first general attribute data to be processed and general game consumption data to be processed corresponding to the target player from the log data, and adjusting the first general attribute data to be processed and the general game consumption data to be processed into the first general attribute data in a standard format and the general game consumption data in a standard format according to a standard format.

In a second aspect, an embodiment of the present application provides a message pushing apparatus, including:

the acquisition module is used for acquiring first general attribute data of a target player in a target game and general game consumption data of the target player;

and the calculation module is used for inputting the first general attribute data and the general game consumption data into a trained prediction model to obtain the churn probability and the payment probability of the target player in the target game.

In a third aspect, an embodiment of the present application provides a computer device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the steps of the above method when executing the computer program.

In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, performs the steps of the above method.

The player prediction method of the online game, which is provided by the embodiment of the application, comprises the steps of firstly, acquiring first general attribute data of a target player in a target game and general game consumption data of the target player; and then, inputting the first general attribute data and the general game consumption data into a trained prediction model to obtain the churn probability and the payment probability of the target player in the target game.

In a certain embodiment, the method provided by the application obtains various general data influencing the churn probability and the payment probability through the first general attribute data and the general game consumption data of the target player, and realizes player prediction of different online games through the two general data, namely the first general attribute data and the general game consumption data. The first general attribute data and the general game consumption data are input into a trained prediction model at the same time, so that the loss probability and the payment probability of the target player can be predicted by simultaneous calculation, the loss probability and the payment probability of the target player can be predicted at the same time, and the prediction efficiency of the target player is improved.

In order to make the aforementioned objects, features and advantages of the present application more comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained from the drawings without inventive effort.

FIG. 1 is a flowchart illustrating a player prediction method for a network game according to an embodiment of the present disclosure;

FIG. 2 is a schematic flow chart illustrating a method for calculating attrition probabilities and payment probabilities according to an embodiment of the present disclosure;

FIG. 3 is a schematic structural diagram of a player prediction device of an online game according to an embodiment of the present disclosure;

fig. 4 is a schematic structural diagram of a computer device according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all the embodiments. The components of the embodiments of the present application, generally described and illustrated in the figures herein, can be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present application, presented in the accompanying drawings, is not intended to limit the scope of the claimed application, but is merely representative of selected embodiments of the application. All other embodiments, which can be derived by a person skilled in the art from the embodiments of the present application without making any creative effort, shall fall within the protection scope of the present application.

In the prior art, in order to retain players of the online game, game developers predict a part of users that the players may lose according to the running data of the online game, and further perform adaptive adjustment on the game. However, when a player is predicted at present, the used prediction data is basically the data unique to the online game played by the predicted player after being analyzed by experts, and the prediction result is possibly closer to the situation of the online game by using the data unique to the online game for prediction, so that the prediction method is not well adapted to other online games. In addition, in the existing scheme, the skill performs single prediction on the loss probability or the payment probability of the game player, and the prediction efficiency is low.

The embodiment of the application provides a player prediction method for an online game, as shown in fig. 1, comprising the following steps:

step S101, acquiring first general attribute data of a target player in a target game and general game consumption data of the target player;

and S102, inputting the first general attribute data and the general game consumption data into the trained prediction model to obtain the churn probability and the payment probability of the target player in the target game.

In the embodiment of the present application, the target game may be any game in which a player can be predicted. The target player may be a player who has used a target game, the target player is determined according to preset rules, and a game developer needs to predict the churn probability and the payment probability of a more active player, so that when determining the target player to be predicted, the target player may be screened according to an active status identifier, where the active status identifier may be a time length for logging in the game in a past period of time (for example, the logging time is 3 days in a past week). The first generic attribute data of the target player may be basic information related to the target player, and the first generic attribute data may include any one or more of the following data: occupation, gender, player level, VIP level, online duration, character creation time, character server, character experience, daily charge amount, cumulative charge amount, daily cost amount, cumulative cost amount, and game character ID number of the target player. Wherein the occupation of the target player refers to the occupation of the target player in real life, such as a doctor. The sex of the target player refers to the sex of the target player in real life, e.g., a man. The target player's online time period is the time period, e.g., 10 hours, that the target player logs into the target game during the day. The target player's game character ID number is a unique identification of the target game character that the target player controls in the game, e.g., Danji _ 1213. The character creation time is a time value at which the target player acquires the game character from the beginning, and if the target player acquires the roban game character in 2019, 10 and 31 days, the creation time of the roban is 2019, 10 and 31 days. The server where the character is located is a server where the game character controls the game character during the game. The character experience value represents experience obtained in the game character process, and the grade of the game character can be improved after the character experience value reaches a certain degree. The daily recharge amount refers to an amount that the target player charges to the target game on the day. The accumulated recharge amount refers to the amount of money charged to the target game by the target player from the time of first logging in the target game to the current moment. The daily payout amount refers to an amount of money that the target player has paid in the target game on the day. The accumulated amount of consumption refers to the amount of consumption in the target game by the target player until the current moment since the target player first logs in the target game. The player rating is a rating determined based on the experience value accumulated by the target player in the target game. The VIP level is a level determined according to a charge amount of the target player to charge the target game. As can be seen from the above, the first common attribute data is data that is commonly present in most network games and can be easily acquired.

The general game consumption data may be data for a player to purchase items in cash in a game, or data for a player to control an equivalent exchange between a virtual character and other virtual characters, and may include any one or more of the following: the method comprises the following steps of purchasing prop name data of the props, purchasing money data of the props, purchasing time data of the props and purchasing time interval data of the props. Wherein, the property name data of the purchased property refers to the property name of the property purchased by the target player in the game mall, such as butcher dragon knife; or the object player controls the object name of the object exchanged with other game characters, such as butcher sword. The amount data of the purchased prop refers to the cash amount spent on purchasing the prop, such as 10 yuan; alternatively, the target game character is equivalently exchanged with other game characters for a virtual amount, such as 100 dollars. The time data of purchasing the prop refers to the time of purchasing the prop, such as 10 months, 02 # and 15 click; alternatively, the time at which the target game character equivalently exchanges with other game characters, for example, 10/month No. 02 and No. 15 click. The time interval data of purchasing the item refers to the time interval of the time when the target player purchases the item twice, and for example, the time interval is 2 hours when the time of the first purchase is 10 months 02 # 15 and the time of the second purchase is 10 months 02 # 17. The target game character is a game character controlled by the target player, and the other virtual characters are virtual characters controlled by the other players. As can be seen from the above, the general game consumption data is data that is ubiquitous in most online games, and is easily accessible.

The first general attribute data and the general game consumption data generate new data every day, if the first general attribute data and the general game consumption data related to the target player in the target game are acquired, the data volume is too much, and long-term historical data may not have a great effect on the prediction effect, so that when the first general attribute data and the general game consumption data are acquired, the first general attribute data and the general game consumption data which need to be acquired can be determined through effective data time. The valid data time is used to limit the range of acquiring the first general attribute data and the general game consumption data, and the valid data time may be the past week, 10 days, or the like.

The first general attribute data and the general game consumption data are general characteristic data about the player extracted from a plurality of kinds of data, and the data structures of the data included in the first general attribute data and the general game consumption data are different, and therefore, the first general attribute data and the general game consumption data constitute multi-source heterogeneous data. The first general attribute data and the general game consumption data can be stored in the HIVE distributed engine, the HIVE distributed engine can process the data, and format adjustment can be performed on the first general attribute data and the general game consumption data with different data structures to obtain data with standard format.

In step S101, after receiving the prediction request (the prediction request carries the active state identifier and the valid data time), a target player is determined from players of the target game according to the active state identifier, and then the first generic attribute data and the generic game consumption data within the valid data time are obtained from data related to the target player. The first general attribute data is the most basic information of the target player, the general game consumption data is the data consumed by the target player in the target game, and the two data are data in most games and are easy to acquire.

In step S102, the prediction model is used to calculate the churn probability and the payout probability of the target player. The attrition probability may be the probability that the target player may stop using the target game and the payout probability may be the probability that the target player may trade in the target game. The higher the churn probability, the more likely the target player will stop using the target game, and the lower the churn probability, the more likely the target player will continue using the target game. The higher the probability of a payout, the more likely the target player is to trade in the target game, and the lower the probability of a payout, the more likely the target player is not to trade in the target game. The paying probability is inversely proportional to the churn probability, the smaller the paying probability is, the more unwilling the target player is to invest for the target game, and further, the bigger the churn probability of the target player is; a higher probability of payout indicates that the target player is more willing to invest in the target game, and thus, the target player's churn probability is lower.

In specific implementation, the first general attribute data and the general game consumption data are input into a trained prediction model, and two values of the churn probability and the payment probability of the target player can be calculated. And calculating two different values by using a trained prediction model, wherein the trained prediction model is a multi-task learning model. Two prediction results are obtained by using one prediction model at the same time, instead of separate calculation of one prediction result and one prediction result, so that the calculation time is saved, and the prediction efficiency of the target player is improved.

In the embodiment of the application, the first general attribute data and the general game consumption data of the target player are obtained in the two steps, various kinds of general data which can influence the loss probability and the payment probability are obtained and are data which are possessed by most games, then the loss probability and the payment probability of the target player are predicted according to the first general attribute data and the general game consumption data, the general data do not need to be analyzed by professional technicians, through the prediction method, the method can be better used for predicting players of various different online games, and the expansibility of the method is improved by utilizing the general data. In the scheme, the first general attribute data and the general game consumption data are input into a trained prediction model, two values of the loss probability and the consumption probability can be obtained at the same time, and the calculation is carried out instead of one value, so that the calculation efficiency is improved.

In this application, the first general attribute data and the general game consumption data are data generated within a valid data time, the data size of the data is relatively large, the features that can be extracted are various, and in order to make the extracted features more accurate, as shown in fig. 2, step S102 includes:

s1021, inputting the occupation, the gender, the online time and the game role ID number in the first general attribute data into a first characteristic evaluation model to obtain a first evaluation result related to the first general attribute data;

s1022, inputting prop name data for purchasing props, sum data for purchasing props, time data for purchasing props and time interval data for purchasing props in the general game consumption data into different evaluation models respectively to obtain evaluation results corresponding to each data;

and S1023, calculating the loss probability and the payment probability of the target player in the target game according to the evaluation results corresponding to the item name data for purchasing the items, the money amount data for purchasing the items, the time data for purchasing the items and the time interval data for purchasing the items and the first evaluation result corresponding to the first general attribute data.

In step S1021, a first feature evaluation model is used for feature evaluation of the first generic attribute data, and the first feature evaluation model may be a convolution model (e.g., a one-dimensional convolution model). The occupation, sex, and game character ID number in the first general attribute data are not substantially changed with time, and only the online time length in the first general attribute data may be changed with time, so that the data amount of the first general attribute data is small, and the first evaluation result on the first general attribute data can be calculated by inputting all the data in the first general attribute data to the first feature evaluation model.

Specifically, after the occupation, the gender, the online time and the game character ID number in the first general attribute data are all input into the first feature evaluation model, the first feature evaluation model extracts features of the occupation, the gender, the online time and the game character ID number to obtain a feature matrix, and the feature matrix is a first evaluation result of the first feature evaluation model.

In step S1022, since each piece of general game consumption data changes with time, a large amount of data is generated in a period of time, and if a large amount of disordered data is placed in a feature extraction model for feature extraction, the extracted features may be relatively single or may have a low accuracy. In order to improve the accuracy of feature extraction, prop name data of prop purchasing, sum data of prop purchasing, prop purchasing time data and prop purchasing time interval data in the general game consumption data are respectively input into different feature evaluation models, and features corresponding to each data are extracted, so that the extracted features are accurate, and further, an evaluation result corresponding to each data is accurate.

Specifically, for each item of property name data of property purchase, amount data of property purchase, time data of property purchase, and time interval data of property purchase in the general game consumption data, feature extraction may be performed by using a corresponding LSTM (Long Short-Term Memory) model, respectively, to obtain a corresponding evaluation result. The prop name data of the purchased props utilize the first LSTM model to obtain a second evaluation result; using the second LSTM model to obtain a third evaluation result according to the amount data of the purchased props; time data of the purchased props are obtained by using a third LSTM model to obtain a fourth evaluation result; and obtaining a fifth evaluation result by using the fourth LSTM model according to the time interval data of the purchased props.

In step S1023, the churn probability and the payout probability of the target player can be calculated based on the evaluation result of each data.

Specifically, a first evaluation result, a second evaluation result, a third evaluation result, a fourth evaluation result and a fifth evaluation result are spliced to obtain a spliced evaluation result, then, an Attention mechanism in a prediction model is used for extracting important features, and the prediction model calculates the loss probability and the payment probability through the extracted important features.

The data corresponding to the target game further includes second general attribute data corresponding to the target player, where the second general attribute data may be related data for describing a game character controlled by the target player, and the second general attribute data is also relatively common data, and is data provided in most network games, and the second general attribute data may include any one or more of the following data: name data of a game character, equipment data of a game character, and skill data of a game character. The name data of the game character refers to a character corresponding to the target game character in the target game, such as Danji. The equipment data of the game character may be the number of equipment currently owned by the target player-controlled target game character in the target game, the grade of the equipment, etc., e.g., the target character has 2 butcher knives in total, and the grade of one butcher knife is grade 11 and the grade of the other butcher knife is grade 8. The skill data of the game character may be a skill name, a level, etc. of a skill that can be released in the target game by the target game character controlled by the target player, for example, the skill is a sun shelter, and the level is 8. Predicting the churn probability and the payout probability of the target player using the augmented second generic attribute data, further comprising:

103, acquiring second general attribute data of the target player in the target game;

step S102, comprising:

and step 1027, inputting the first general attribute data, the second general attribute data and the general game consumption data into the trained prediction model to obtain the churn probability and the payment probability of the target player in the target game.

In step 103, in the present embodiment, in addition to the first general attribute data and the general game consumption data of the target player, the second general attribute data needs to be acquired. The second general attribute data may be basic data of a game character played by the target player, and is data that is provided in most games. The second generic attribute data may also be stored in the HIVE distribution engine, and the data format of the second generic attribute data may also be different from the data format of the first generic attribute data and the generic game consumption data, and therefore, the second generic attribute data is also data that requires the HIVE distribution engine to perform data processing so as to adjust the second generic attribute data to a standard format.

In step 1027, in order to improve the prediction accuracy of the churn probability and the payout probability of the target player, the first general attribute data, the second general attribute data and the general game consumption data can be input into the trained prediction model at the same time to calculate the churn probability and the payout probability of the target player. The data about the second general attribute, namely the types of the data used for predicting the churn probability and the payment probability of the target player are increased, and the accuracy of prediction of the churn probability and the payment probability is improved.

In order to improve the accuracy of extracting the features, the following steps can be adopted to respectively extract the features of different data:

step 10271, inputting occupation, gender, online duration and game role ID number in the first general attribute data, and name data, equipment data and skill data of the game role in the second general attribute data into the first characteristic evaluation model to obtain a first evaluation result related to the first general attribute data;

step 10272, inputting item name data of item purchasing, money amount data of item purchasing, item purchasing time data and item purchasing time interval data in the general game consumption data into different evaluation models respectively to obtain an evaluation result corresponding to each data;

and step 10273, calculating the churn probability and the payment probability of the target player in the target game according to the evaluation results corresponding to the item name data of the purchased items, the money amount data of the purchased items, the time data of the purchased items and the time interval data of the purchased items, and the first evaluation result corresponding to the first general attribute data.

In step 10271, since the name data of the game character, the equipment data of the game character, and the skill data of the game character in the second general attribute data are data with a small data amount, the first evaluation result can be calculated together with the occupation, sex, online time, and game character ID number in the first general attribute data by inputting them to the first feature evaluation model.

The detailed discussion of steps 10272-10273 can refer to the discussion of steps 1022-1023 above.

The trained model is obtained through training, in order to obtain accurate loss probability and payment probability, a large amount of data needs to be used for training the model to be trained, and the trained prediction model is obtained through the following steps:

step 104, acquiring a training sample set; the sample set comprises at least one training sample, and each training sample comprises first training data and second training data; the first training data includes first generic attribute data of the player and generic game consumption data of the player; the second training data includes attrition labels and bet labels for the player;

and 105, aiming at each training sample, simultaneously inputting the first training data and the second training data in the training sample to the prediction model to be trained, and training the prediction model to be trained.

In step 104, in order to improve the accuracy of the model, a large number of training samples are set in the sample set, each training sample is the first general attribute data and the general game consumption data corresponding to one player, and certainly, there is also an attrition tag indicating whether the player is attrited, the attrition tag may be represented by a preset character, such as a number, a letter, etc., in this application, the attrition tag indicating that the player has attrited is 1, and the attrition tag not attrited is 0. And a payment tag indicating whether the player pays, where the payment tag may be represented by a preset character, such as a number, a letter, etc., and in this application, the payment tag indicating that the player paid is 1, and the payment tag not paid is 0.

In step 105, for each training sample, the prediction model to be trained is calculated once, the first general attribute data, the general game consumption data, the churn label and the payment label of the player are simultaneously input into the prediction model to be trained, and the prediction model to be trained is adjusted in the training process.

The process of adjusting the predictive model to be trained includes:

step 1051, aiming at each training sample, according to the comparison result between the calculation result of the first training data and the second training data, adjusting the weight value of each feature in the prediction model to be trained so as to train the prediction model to be trained until the comparison result between the calculation result of the positive sample and the negative sample reaches the preset precision.

In step 1051, after the first general attribute data and the general game consumption data of the player are input into the prediction model to be trained, the prediction model to be trained outputs the calculation results of the churn probability and the payment probability, compares the calculation results of the churn probability with the churn labels in the second training data, compares the calculation results of the payment probability with the payment labels in the second training data, and adjusts the weight value of each feature in the prediction model to be trained according to the comparison results. The weight value of each feature is generally adjusted by a random gradient descent method. And training the prediction model to be trained is realized in the process of adjusting the weight value of each characteristic value, and the training is stopped until the comparison result between the positive sample calculation result and the negative sample of the prediction model to be trained reaches the preset precision. The preset precision is set manually.

In this scheme, after obtaining the churn probability and the payout probability, the game developer can take corresponding measures according to the two values to reduce the churn of the game player, and therefore, this scheme further includes:

and 106, determining to push a message related to the target game to the target player according to the attrition probability and the payment probability.

In step 106, the message regarding the target game may be a message for improving the interest level of the target player in the target game, and the message regarding the target game may include any one or more of the following messages: offer strategies for the target game and improvement strategies for the target game. The preferential strategy related to the target game can be a preferential strategy aiming at commodities, activities and the like in the game and is a strategy related to the consumption condition of the target game, and the preferential strategy can comprise any one or more of the following strategies: providing lottery activity, providing game privileges, upscaling property discounts. Wherein providing a lottery event may be a means of soliciting players for more interest with less effort, e.g., not spending money to obtain items. Providing game privileges may be to have a ranked user unlock a higher level game scenario, e.g., a 22 level user tries to play in a 33 level scenario for half an hour. The discount for the prop may be to reduce the selling price of the prop in the target game, for example, 55 pistol with a present price of 20. The improvement strategy for the target game may be an improvement of the target game itself, and in particular, the improvement strategy for the target game may include any one or more of the following strategies: optimizing game pages, increasing game level and improving game fluency speed. Optimizing the game page refers to adjusting graphics, colors and the like in the game interface, for example, adding a background picture in the game interface. Adding a game level may be setting a game level for the target group, such as setting an adult mode level, a child mode level, and the like. The improvement of the game fluency speed can be the improvement of the running efficiency of the game, for example, the reduction of the cache of the game and the like.

In the specific implementation, after the churn probability and the payment probability of the target player are determined, whether the target player wants to stop using the target game can be predicted, and for the target player who wants to stop using the target game, relevant information about the target game can be pushed to the target player to attract the target player, so that the churn of the target player is reduced, and the number of players using the target game is increased.

In an embodiment of the present application, a more specific method for determining to push a message to a target player is provided, step 106, comprising:

step 1061, if the attrition probability of the target player is less than the first attrition probability preset value and the payment probability is less than the first payment probability preset value, determining to push a preferential strategy to the target player;

step 1062, if the churn probability of the target player is greater than the first churn probability preset value and less than the second churn probability preset value, and the payment probability is less than the first payment probability preset value and less than the second payment probability preset value, determining to push two preferential strategies to the target player at the same time;

and step 1063, if the attrition probability of the target player is greater than the second attrition probability preset value, and the payment probability is greater than the second payment probability preset value, determining to simultaneously push three preferential strategies to the target player.

In step 1061, the first attrition probability preset value and the first payment probability preset value are both manually specified, and the first attrition probability preset value and the first payment probability preset value may be equal or unequal.

In a specific implementation, if the churn probability of the target player is smaller than the first churn probability preset value and the payout probability is smaller than the first payout probability preset value, this indicates that the target player has a lower churn probability and a lower consumption probability, and therefore, the target player may be a player with higher stability, and the target players may be nursed by one or more preferential policies, so that the target player is more stable.

In the step 1062, the second predetermined loss probability value and the second predetermined payment probability value are both manually specified, and the second predetermined loss probability value and the second predetermined payment probability value may be equal or unequal. The second loss probability preset value is larger than the first loss probability preset value, and the second payment probability preset value is larger than the first payment probability preset value.

In specific implementation, if the churn probability of the target player is greater than the first churn probability preset value and less than the second churn probability preset value, and the payout probability is less than the first payout probability preset value and less than the second payout probability preset value, this indicates that the target player has a high possibility of churn and a high possibility of consumption, and this part of users can bring high income for the target game, so that this part of target players can be saved by two or more preferential policies, the interest of this part of players in the target game is improved, and the churn of the target player is reduced.

In the step 1063, if the churn probability of the target player is greater than the second churn probability preset value and the payout probability is greater than the second payout probability preset value, this indicates that the possibility of churn and payout of the target player is high, and this part of users can bring higher profit to the target game, so that this part of target players can be saved by three or more preferential policies, the interest of this part of players in the target game is increased, and the churn of the target player is reduced.

Besides providing the target player with the benefits, the visual effect and the use effect brought by the game to the target player can be changed, so as to enhance the interest of the player from multiple aspects and reduce the loss of the player, and the step 106 comprises the following steps:

and step 1064, if the attrition probability of the target player is greater than the third attrition probability preset value, and the payment probability is greater than the third payment probability preset value, determining to push a discount strategy about the target game and an improvement strategy about the target game to the target player.

In step 1064, the third predetermined value of the loss probability and the third predetermined value of the fee probability are both manually specified, and the third predetermined value of the loss probability and the third predetermined value of the fee probability may be equal or unequal.

In specific implementation, if the churn probability of the target player is greater than the third churn probability preset value and the payout probability is greater than the third payout probability preset value, this indicates that the churn probability and the payout probability of the target player are high, and in order to improve the interest of the target player in the target game, the target game itself can be improved while a preference strategy is provided for the target player, so that the visual effect of the target game is improved, the trial play range of the target game is improved, and the performance of the target game is improved. The improvement of the target game from multiple aspects and the attraction of the target player from multiple aspects can reduce the loss of the target player.

The data corresponding to the target player is acquired from the database of the target game, and step S101 includes:

step 1011, obtaining the log data corresponding to the target player from the database of the target game;

step 1012, the first general attribute data to be processed and the general game consumption data to be processed corresponding to the target player are screened out from the log data, and the first general attribute data to be processed and the general game consumption data to be processed are adjusted to the first general attribute data in the standard format and the general game consumption data in the standard format according to the standard format.

In step 1011, the database of the target game is used to store log data generated during the running of the target game, and data stored in the database is generally stored in units of days. The log data may include first general attribute data, general game consumption data, and more data, and the data in the log is arranged according to a preset rule.

In the above step 1012, the standard format refers to one-dimensional convolution of each data in the first common attribute data and hash processing of each data in the common game consumption data.

In specific implementation, the data in the log data includes multiple types, and therefore, the first general attribute data and the general game consumption data need to be screened from the log data according to a preset rule, but the data directly screened from the log data does not meet the input standard of the prediction model to be trained, and therefore, in order to accurately predict the attrition probability and the payment probability, the first general attribute data to be processed and the general game consumption data to be processed, which are screened from the log data, need to be adjusted according to a standard format to obtain the first general attribute data and the general game consumption data in the standard format. The attrition probability and the payment probability can be accurately predicted by using the first general attribute data and the general game consumption data in the standard format.

As shown in fig. 3, an embodiment of the present application provides a player prediction apparatus for a network game, including:

an obtaining module 301, configured to obtain first general attribute data of a target player in a target game and general game consumption data of the target player;

and the calculating module 302 is configured to input the first general attribute data and the general game consumption data into the trained prediction model, so as to obtain the churn probability and the payment probability of the target player in the target game.

Optionally, the first generic attribute data includes any one or more of the following data:

occupation, gender, player level, VIP level, online duration, character creation time, server where the character resides, character experience value, daily charge amount, cumulative charge amount, daily cost amount, cumulative cost amount, and game character ID number of the target player.

Optionally, the general game consumption data comprises any one or more of the following:

Optionally, the calculating module 302 includes:

the first calculation unit is used for inputting the occupation, the gender, the online time length and the game role ID number in the first general attribute data into the first characteristic evaluation model to obtain a first evaluation result related to the first general attribute data;

the second calculation unit is used for respectively inputting prop name data for purchasing props, money data for purchasing props, time data for purchasing props and time interval data for purchasing props in the general game consumption data into different evaluation models to obtain evaluation results corresponding to each data;

and the third calculating unit is used for calculating the loss probability and the payment probability of the target player in the target game according to the evaluation results corresponding to the item name data for purchasing the item, the money amount data for purchasing the item, the time data for purchasing the item and the time interval data for purchasing the item and the first evaluation result corresponding to the first general attribute data.

Optionally, the calculating module 302 includes:

and the fourth calculating unit is used for inputting the first general attribute data, the second general attribute data and the general game consumption data into the trained prediction model to obtain the churn probability and the payment probability of the target player in the target game.

Optionally, the apparatus further comprises:

the sample acquisition module is used for acquiring a training sample set; the sample set comprises at least one training sample, and each training sample comprises first training data and second training data; the first training data includes first generic attribute data of the player and generic game consumption data of the player; the second training data includes attrition labels and bet labels for the player;

and the training module is used for inputting the first training data and the second training data in the training samples to the prediction model to be trained simultaneously aiming at each training sample, and training the prediction model to be trained.

Optionally, the training module includes:

and aiming at each training sample, according to the comparison result between the calculation result of the first training data and the second training data, adjusting the weight value of each feature in the prediction model to be trained so as to train the prediction model to be trained until the comparison result between the calculation result of the positive sample and the negative sample reaches the preset precision.

Optionally, the apparatus further comprises:

and the pushing module is used for determining to push the information about the target game to the target player according to the attrition probability and the payment probability.

Optionally, the message regarding the target game includes any one or more of the following messages:

offer strategies for the target game and improvement strategies for the target game.

Optionally, the improvement policy comprises any one or more of the following policies:

optimizing game pages, increasing game level and improving game fluency speed.

Optionally, the pushing module includes:

the first pushing unit is used for determining to push a discount strategy to the target player if the attrition probability of the target player is smaller than a first attrition probability preset value and the payment probability is smaller than a first payment probability preset value;

the second pushing unit is used for determining that two preferential strategies are pushed to the target player at the same time if the churn probability of the target player is greater than the first churn probability preset value and smaller than the second churn probability preset value, and the payment probability is smaller than the first payment probability preset value and smaller than the second payment probability preset value;

and the third pushing unit is used for determining that three preferential strategies are pushed to the target player at the same time if the attrition probability of the target player is greater than the second attrition probability preset value and the payment probability is greater than the second payment probability preset value.

Optionally, the pushing module includes:

and the fourth pushing unit is used for determining to push a discount strategy about the target game and an improvement strategy about the target game to the target player if the churn probability of the target player is greater than the third churn probability preset value and the payment probability is greater than the third payment probability preset value.

Optionally, the obtaining module 101 includes:

the acquisition unit is used for acquiring the log data corresponding to the target player from the database of the target game;

and the adjusting unit is used for screening out the first general attribute data to be processed and the general game consumption data to be processed corresponding to the target player from the log data, and adjusting the first general attribute data to be processed and the general game consumption data to be processed into the first general attribute data in the standard format and the general game consumption data in the standard format according to the standard format.

Corresponding to the player prediction method of the network game in fig. 1, an embodiment of the present application further provides a computer device 400, as shown in fig. 4, the device includes a memory 401, a processor 402, and a computer program stored on the memory 401 and executable on the processor 402, wherein the processor 402 implements the player prediction method of the network game when executing the computer program.

Specifically, the memory 401 and the processor 402 can be general memories and processors, which are not limited in particular, and when the processor 402 runs a computer program stored in the memory 401, the player prediction method of the online game can be executed, so that the problems that the churn probability and the payment probability cannot be predicted for one game at the same time and different games cannot be well adapted in the prior art are solved.

Corresponding to the player prediction method of the network game in fig. 1, an embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the steps of the player prediction method of the network game.

Specifically, the storage medium can be a universal storage medium, such as a mobile disk, a hard disk and the like, when a computer program on the storage medium is run, the player prediction method of the online game can be executed, the problems that the churn probability and the payment probability cannot be predicted simultaneously for one game and different games cannot be well adapted in the prior art are solved, the application obtains a plurality of common data which can influence the churn probability and the payment probability and are data possessed by most games by obtaining first universal attribute data and universal game consumption data of a target player, then predicts the churn probability and the payment probability of the target player according to the first universal attribute data and the universal game consumption data, the common data does not need professional technicians to analyze the common data, and the prediction method can be better used for predicting players of a plurality of different online games, the utilization of common data improves the expansibility of the method. In the scheme, the first general attribute data and the general game consumption data are input into a trained prediction model, two values of the loss probability and the consumption probability can be obtained at the same time, and the calculation is carried out instead of one value, so that the calculation efficiency is improved.

In the embodiments provided in the present application, it should be understood that the disclosed method and apparatus may be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one logical division, and there may be other divisions when actually implemented, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or units through some communication interfaces, and may be in an electrical, mechanical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.

It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus once an item is defined in one figure, it need not be further defined and explained in subsequent figures, and moreover, the terms "first", "second", "third", etc. are used merely to distinguish one description from another and are not to be construed as indicating or implying relative importance.

Finally, it should be noted that: the above-mentioned embodiments are only specific embodiments of the present application, and are used for illustrating the technical solutions of the present application, but not limiting the same, and the scope of the present application is not limited thereto, and although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: any person skilled in the art can modify or easily conceive the technical solutions described in the foregoing embodiments or equivalent substitutes for some technical features within the technical scope disclosed in the present application; such modifications, changes or substitutions do not depart from the spirit and scope of the present disclosure, which should be construed in light of the above teachings. Are intended to be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A player prediction method for a network game, comprising:

2. The method of claim 1, wherein the first generic attribute data comprises any one or more of:

occupation, gender, player grade, VIP grade, online time, character creation time, a server where the character is located, a character experience value, a daily recharge amount, an accumulated recharge amount, a daily consume amount, an accumulated consume amount and a game character ID number of the target player.

3. The method of claim 1, wherein the generic game consumption data comprises any one or more of the following:

4. The method of claim 3, wherein inputting the first generic attribute data and the generic game consumption data into a trained predictive model to derive the churn probability and the payout probability for the target player in the target game comprises:

5. The method of claim 1, wherein inputting the first generic attribute data and the generic game consumption data into a trained predictive model to derive the churn probability and the payout probability for the target player in the target game comprises:

6. The method according to claim 5, wherein the second generic attribute data comprises any one or more of:

7. The method of claim 1, wherein the predictive model is trained by:

8. The method according to claim 7, wherein for each training sample, the first training data and the second training data in the training sample are simultaneously input to a predictive model to be trained, and the training of the predictive model to be trained comprises:

9. The method of claim 1, further comprising:

10. The method of claim 9, wherein the message regarding the target game comprises any one or more of:

11. The method of claim 10, wherein the offer policy comprises any one or more of the following policies:

12. The method of claim 10, wherein the improvement policy comprises any one or more of the following policies:

optimizing game pages, increasing game level and improving game fluency speed.

13. The method of claim 11, wherein determining to push a message to the target player regarding the target game based on the attrition probability and the payout probability comprises:

14. The method of claim 10, wherein determining to push a message to the target player regarding the target game based on the attrition probability and the payout probability comprises:

15. The method of claim 1, wherein obtaining first generic attribute data of a target player in a target game and generic game consumption data of the target player comprises:

16. A player prediction apparatus for a network game, comprising:

17. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps of the method of any of the preceding claims 1-15 when executing the computer program.

18. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method according to any one of the claims 1-15.